Recent Insights
Coding ability: In Codex, GPT 5.2-codex-high is the strongest. It is suitable for writing more complex products or finding hard-to-solve bugs. It can solve several bugs that others cannot. The downside is that it is very slow, so it shouldn't be used for small changes. Making a small adjustment can take as long as a week of reading "A Mortal's Journey to Immortality."
Next in coding ability is Claude Opus 4.5. Its quality and capability are second only to the top one. However, it is much faster and can be used in Antigravity.
Gemini 3 Pro. It is the strongest in frontend development; I delegate all my frontend tasks to it, and its aesthetic sense is generally good. Besides coding, it is highly recommended for other scenarios as well. It handles daily conversations and connects to the Google ecosystem. Google is very generous, providing ample resources; applying for an educational qualification in the U.S. allows free use for one year.
Claude Sonnet 4.5 is quite awkward. I haven't used it for a while. Its capabilities are not as strong as the first three. If using the API, the price is still quite high. If it's a monthly subscription, Opus 4.5 is more recommended.
A cost-effective choice is GLM 4.7. Its coding ability is decent, and their model updates quickly. Importantly, the price is affordable, costing less than 20 RMB a month. It is basically underutilized. It can be used for small modifications, etc.
Don't put too much trust in AI rankings, nor in those self-media outlets. Whenever a new one comes out, they claim it's the world's best. It's unclear whether they are being dishonest or simply don't understand. It wastes others' time and energy.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
